Deep-STP: a deep learning-based approach to predict snake toxin proteins by using word embeddings.

Area of Science:

• Biochemistry
• Computational Biology
• Pharmacology

Background:

• Snake venom contains toxic proteins impacting circulatory and nervous systems.
• These toxins show therapeutic potential for cardiovascular and neurological diseases.
• Traditional biochemical methods for identifying venom proteins are costly and time-consuming.

Purpose of the Study:

• To develop a sequence-based computational method for recognizing snake toxin proteins.
• To leverage artificial intelligence for large-scale screening of venom proteins.
• To facilitate drug development for cardiovascular and nervous system diseases.

Main Methods:

• Utilized sequence-based computational methods.
• Employed feature descriptors: g-gap, natural vector, and word 2 vector.
• Applied analysis of variance (ANOVA), gradient-boost decision tree algorithm (GBDT), and incremental feature selection (IFS) for feature optimization.
• Trained a deep learning model with optimized features.

Main Results:

• Achieved 82.00% accuracy in 10-fold cross-validation.
• Reached 81.14% accuracy on independent data.
• Demonstrated excellent prediction performance and robustness of the developed model.

Conclusions:

• The developed AI model provides an efficient computational approach for snake toxin protein identification.
• This method can accelerate the discovery of novel drug candidates from snake venom.
• The study highlights the potential of AI in advancing venom-based drug development.